Chapter Summary

Chapter Summary

Key Points

• 1.

  The Gaussian relay channel: capacity to within 0.5 bits. For the Gaussian relay channel $Y = X + X_r + Z$, $Y_r = X + Z_r$, DF achieves capacity when $N_r \leq N$ (degraded case) via coherent combining. CF achieves within 0.5 bits of the cut-set bound for all parameter regimes — a universal constant-gap result. The gap comes from independent (vs. correlated) inputs, costing at most $\frac{1}{2}\log 2 = 0.5$ bits.

• 2.

  DF vs. CF: it depends on the link qualities. DF excels when the source-relay link is strong ($N_r$ small); CF excels when the relay-destination link is strong. Neither strategy uniformly dominates. Practical systems should adaptively select based on measured link quality.

• 3.

  Gupta-Kumar: multi-hop routing does not scale. For $n$ nodes in a unit square with multi-hop routing, the per-node throughput scales as $\Theta(1/\sqrt{n\log n})$ — it vanishes as the network grows. Aggregate interference is the fundamental bottleneck.

• 4.

  Hierarchical cooperation achieves linear scaling. Özgür, Lévêque, and Tse showed that cooperative communication via virtual MIMO arrays achieves $\Theta(n)$ total throughput for path-loss exponent $\alpha = 2$. Cooperation converts interference into useful signal.

• 5.

  Noisy network coding: a universal relay strategy. NNC generalizes CF to arbitrary networks: each relay compresses independently, and the destination jointly decodes. NNC achieves within $(K-1)/2$ bits of the cut-set bound for any $K$-node Gaussian network — a constant gap independent of channel parameters.